A while ago (r4453, early May), Vikram Garg and I added the patch_reuse option to PatchRecoveryErrorEstimator, which tells the estimator to use the same patch to generate error estimates for every element within it, rather than generating a new patch and new postprocessed patch solution for each element.
We made this option default to true and didn't bother advertising the change because it runs an order of magnitude faster, it is supposed to have the same asymptotic effectivity behavior (up to a constant), and it gave nearly identical adaptivity performance in our tests. However, it turns out that that constant matters, and the new default PREE gives visibly inferior adaptivity performance in my tests using a different application with a more complex and fragile formulation. So a very belated heads-up: anyone using patch recovery (or adjoint residual, which by default depends on patch recovery) error estimates might want to explore both sides of the reuse option. --- Roy ------------------------------------------------------------------------------ All the data continuously generated in your IT infrastructure contains a definitive record of customers, application performance, security threats, fraudulent activity and more. Splunk takes this data and makes sense of it. Business sense. IT sense. Common sense. http://p.sf.net/sfu/splunk-d2dcopy1 _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
